TikTok Engagement Traces Over Time and Health Risky Behaviors: Combining Data Linkage and Computational Methods
Xinyan Zhao, Chau-Wai Wong
TL;DR
This study addresses how selective engagement with health-related TikTok content relates to vaping and drinking behaviors. It combines survey data with consent-based TikTok traces and applies Deepgram transcription and GPT-4-based valence coding to analyze 13,724 liked videos from 2020–2023, enabling objective, time-resolved links between engagement patterns and self-reported behaviors. Key findings show heterogeneous temporal patterns in liking drinking videos, with some users accelerating engagement over time, and robust associations between the size of liked content and current vaping/drinking behaviors, while fruit/vegetable content shows protective associations; valence effects are limited. The work demonstrates the value of data linkage and computational analysis for ecologically valid health communication research, highlighting methodological advantages and limitations, and suggesting that attention to timeframes and cross-topic engagement is crucial for understanding social media health effects.
Abstract
Digital technologies and social algorithms are revolutionizing the media landscape, altering how we select and consume health information. Extending the selectivity paradigm with research on social media engagement, the convergence perspective, and algorithmic impact, this study investigates how individuals' liked TikTok videos on various health-risk topics are associated with their vaping and drinking behaviors. Methodologically, we relied on data linkage to objectively measure selective engagement on social media, which involves combining survey self-reports with digital traces from TikTok interactions for the consented respondents (n = 166). A computational analysis of 13,724 health-related videos liked by these respondents from 2020 to 2023 was conducted. Our findings indicate that users who initially liked drinking-related content on TikTok are inclined to favor more of such videos over time, with their likes on smoking, drinking, and fruit and vegetable videos influencing their self-reported vaping and drinking behaviors. Our study highlights the methodological value of combining digital traces, computational analysis, and self-reported data for a more objective examination of social media consumption and engagement, as well as a more ecologically valid understanding of social media's behavioral impact.